Skip wrote:
I only use two commands to process my mail:
dspam --debug --stdout --deliver=innocent,spam --mode=teft
--feature=noise,whitelist --user skip
for all incoming email
dspam --mode=teft --source=error --class=spam --user skip
to fix the missed spammies
Of course, I did feed a bunch of spam and ham corpus when I first set
everything up, but why would I still have a few new spam corpusfed
each day?
This table shows the output of dspam_stats at the end of each day
since I set this up.
DP=Daily positives
DN=Daily Negatives
DFP=Daily False Positives
DFN=Daily False Negatives
DSC=Daily Spam Corpusfed
and those are just the change from the previous day's value.
Date TP TN FP FN SC NC SHR HSR OSA DP DN DFP
DFN DSC
3/04/2008 24 23 0 2 210 3194 0.92 0 0.96 24 23 0
2 210
3/05/2008 41 36 0 3 214 3194 0.93 0 0.96 17 13 0
1 4
3/06/2008 75 60 0 5 219 3194 0.94 0 0.96 34 24 0
2 5
3/07/2008 92 79 0 7 221 3194 0.93 0 0.96 17 19 0
2 2
3/08/2008 115 94 0 14 228 3194 0.89 0 0.94 23 15 0
7 7
3/09/2008 167 103 0 29 255 3194 0.85 0 0.90 52 9 0
15 27
3/10/2008 191 110 0 40 264 3194 0.83 0 0.88 24 7 0
11 9
3/11/2008 254 122 0 46 270 3194 0.85 0 0.89 63 12 0
6 6
3/12/2008 288 136 0 53 283 3194 0.84 0 0.89 34 14 0
7 13
3/13/2008 328 148 0 63 290 3194 0.84 0 0.88 40 12 0
10 7
3/14/2008 374 163 0 66 291 3194 0.85 0 0.89 46 15 0
3 1
3/15/2008 431 180 0 74 303 3194 0.85 0 0.89 57 17 0
8 12
3/16/2008 463 194 0 79 309 3194 0.85 0 0.89 32 14 0
5 6
3/17/2008 511 204 0 88 314 3194 0.85 0 0.89 48 10 0
9 5
As you can see, I do get a few new corpusfed spams each day, but I
know I am not using the source=corpus option. Could this be affecting
my performance?
Regards,
Skip
Skip,
If you have the TestConditionalTraining option on, then dspam will
retrain (corpus-feed) the message a few times, which can make that
counter go up.
Also, you do not need to specify --mode or --feature on the command
line, as dspam will automatically pick them up from either dspam.conf,
or from the preferences file.
ALSO, if you are only using dspam --mode=teft --source=error
--class=spam --user skip for retraining, then how do you fix wrongly
classified ham?
--Kyle Johnson!
